Obousměrné generování a validace softwarového kódu a požadavků pomocí AI

Abstract

This thesis addresses the design, implementation, and evaluation of a tool for bidirectional generation and validation between software requirements and source code using advanced artificial intelligence models. The goal is to explore the automation of transformations between structured requirements in the ReqPat and User Stories format, natural language specifications, and corresponding functional code, as well as the reverse transformation from code back to requirements. The tool integrates multiple AI models (OpenAI, Gemini, Claude and Llama) to enable cross-model validation and consistency checks. A key feature is an interactive editor with visual links between requirements and code, enabling traceability and consistency management. The thesis includes the system architecture, validation methodology, and experimental evaluation on selected scenarios. Results demonstrate the potential of large language models in supporting bidirectional software development and highlight current limitations in terms of accuracy, reliability, and explainability.